swiftpy

Minimal SWIFT trigger activity builder for DUNE-HD.

Takes a pickle file of Trigger Primitives (TPs) and produces a DataFrame of Trigger Activities (TAs) using local energy estimation in a fixed time grid + DBSCAN clustering.

Dependencies

pip install numpy pandas numba tqdm

Quickstart

(see test.py)

import pandas as pd
from tamaker import TAMaker
from presets import COLLECTION_CENTRAL

df = pd.read_pickle("my_tps.pkl")

maker = TAMaker(**COLLECTION_CENTRAL)
tas, tps = maker.run(df)

tas.to_pickle("tas.pkl")
tps.to_pickle("tps_annotated.pkl")

For large datasets spread across multiple runs:

tas, tps = maker.run_by_run(df, output_ta="tas.pkl", output_tp="tps.pkl")

Alg. Parts

1. Filter — TPs below peak_adc_cut or tot_cut are dropped.
2. Window — the readout is divided into fixed-width time bins. Each (event, apa, rop, bin) group is summed to give a window ADC total.
3. Flag — windows are flagged:
   • flag=2 (accept) if window ADC > accept_threshold
   • flag=1 (inspect) if window ADC > inspect_threshold
   • flag=0 (reject) otherwise
4. Cluster — inspect windows go through DBSCAN. If the largest cluster exceeds cluster_cut, the flag is promoted to 2, otherwise it becomes 0.
5. Output — only flag-1 and flag-2 windows produce TA rows. The TP DataFrame is returned with TA_id and dbscan_label columns added.

Presets

Name	Plane	inspect thr	accept thr	cluster cut
COLLECTION_CENTRAL	2	15k	55k	15k
COLLECTION_LATERAL	2	15k	130k	17k
INDUCTION_V_CENTRAL	1	3.5k	30k	5k
INDUCTION_U_CENTRAL	0	3.2k	30k	4.5k

You can also pass parameters directly to override anything:

maker = TAMaker(**COLLECTION_CENTRAL, cluster_cut=20e3)

Output columns

tas

Column	Description
event	Event number
apa_id	APA (detector module) index
rop	Readout plane index
window_start	Start tick of the time window
flag	0=rejected, 2=accepted
TA_id	Unique trigger activity ID
total_window_energy	Sum of TP ADC integrals in the window
TP_count	Number of TPs in the window
n_clusters	DBSCAN clusters found (-1 for immediate accepts)
max_cluster_energy	ADC sum of the largest cluster
ch_mean, t_mean	Unweighted centroid of the TP cloud
ch_wmean, t_wmean	Charge-weighted centroid

tps — same as the input TP DataFrame filtered to the selected plane, with TA_id, dbscan_label, and window_start columns added. This can be used for associating TP-TA information